zoukankan      html  css  js  c++  java
  • Flink学习笔记:Operators串烧

    本文为《Flink大数据项目实战》学习笔记,想通过视频系统学习Flink这个最火爆的大数据计算框架的同学,推荐学习课程:

    Flink大数据项目实战:http://t.cn/EJtKhaz

    1. DataStream Transformation

    1.1 DataStream转换关系

     

    上图标识了DataStream不同形态直接的转换关系,也可以看出DataStream主要包含以下几类:

    1.keyby就是按照指定的key分组

    2.window是一种特殊的分组(基于时间)

    3.coGroup

    4.join Join是cogroup 的特例

    5.Connect就是松散联盟,类似于英联邦

    1.2 DataStream

    DataStream 是 Flink 流处理 API 中最核心的数据结构。它代表了一个运行在多个分区上的并行流。

    一个 DataStream 可以从 StreamExecutionEnvironment 通过env.addSource(SourceFunction) 获得。

    1.3 map&flatMap

    含义:数据映射(1进1出和1进n出)

    转换关系:DataStream → DataStream

    使用场景:

    ETL时删减计算过程中不需要的字段

    案例1:

    public class TestMap {

        public static void main(String[] args) throws Exception {

            final StreamExecutionEnvironment env=StreamExecutionEnvironment.getExecutionEnvironment();

            DataStream<Long> input=env.generateSequence(0,10);

            DataStream plusOne=input.map(new MapFunction<Long, Long>() {

                @Override

                public Long map(Long value) throws Exception {

                    System.out.println("--------------------"+value);

                    return value+1;

                }

            });

            plusOne.print();

            env.execute();

        }

    }

    案例2:

    public class TestFlatmap {

        public static void main(String[] args) throws Exception {

            final StreamExecutionEnvironment env=StreamExecutionEnvironment.getExecutionEnvironment();

            DataStream<String> input=env.fromElements(WORDS);

            DataStream<String> wordStream=input.flatMap(new FlatMapFunction<String, String>() {

                @Override

                public void flatMap(String value, Collector<String> out) throws Exception {

                    String[] tokens = value.toLowerCase().split("\W+");

                    for (String token : tokens) {

                        if (token.length() > 0) {

                            out.collect(token);

                        }

                    }

                }

            });

            wordStream.print();

            env.execute();

        }

        public static final String[] WORDS = new String[] {

                "To be, or not to be,--that is the question:--",

                "Whether 'tis nobler in the mind to suffer",

                "The slings and arrows of outrageous fortune",

                "And by opposing end them?--To die,--to sleep,--",

                "Be all my sins remember'd."

        };

    }

     

    如右上图所示,DataStream 各个算子会并行运行,算子之间是数据流分区。如 Source 的第一个并行实例(S1)和 flatMap() 的第一个并行实例(m1)之间就是一个数据流分区。而在 flatMap() 和 map() 之间由于加了 rebalance(),它们之间的数据流分区就有3个子分区(m1的数据流向3个map()实例)。这与 Apache Kafka 是很类似的,把流想象成 Kafka Topic,而一个流分区就表示一个 Topic Partition,流的目标并行算子实例就是 Kafka Consumers。

    1.4 filter

    含义:数据筛选(满足条件event的被筛选出来进行后续处理),根据FliterFunction返回的布尔值来判断是否保留元素,true为保留,false则丢弃 

    转换关系: DataStream → DataStream 

    使用场景:

    过滤脏数据、数据清洗等

     

    案例:

    public class TestFilter {

        public static void main(String[] args) throws Exception {

            final StreamExecutionEnvironment env=StreamExecutionEnvironment.getExecutionEnvironment(); 

            DataStream<Long> input=env.generateSequence(-5,5);

            input.filter(new FilterFunction<Long>() {

                @Override

                public boolean filter(Long value) throws Exception {

                    return value>0;

                }

            }).print();

            env.execute();

        }

    }

    1.5 keyBy

    含义:

    根据指定的key进行分组(逻辑上把DataStream分成若干不相交的分区,key一样的event会被划分到相同的partition,内部采用hash分区来实现)

    转换关系: DataStream → KeyedStream

    限制:

    1.可能会出现数据倾斜,可根据实际情况结合物理分区来解决(后面马上会讲到)

    2.Key的类型限制:

    1)不能是没有覆盖hashCode方法的POJO

    2)不能是数组

    使用场景:

    1.分组(类比SQL中的分组 

    案例:

    public class TestKeyBy {

        public static void main(String[] args) throws Exception {

            //统计各班语文成绩最高分是谁

            final StreamExecutionEnvironment env=StreamExecutionEnvironment.getExecutionEnvironment();

            DataStream<Tuple4<String,String,String,Integer>> input=env.fromElements(TRANSCRIPT);

            KeyedStream<Tuple4<String,String,String,Integer>,Tuple> keyedStream = input.keyBy("f0");

            keyedStream.maxBy("f3").print();

            env.execute();

        }

        public static final Tuple4[] TRANSCRIPT = new Tuple4[] {

                Tuple4.of("class1","张三","语文",100),

                Tuple4.of("class1","李四","语文",78),

                Tuple4.of("class1","王五","语文",99),

                Tuple4.of("class2","赵六","语文",81),

                Tuple4.of("class2","钱七","语文",59),

                Tuple4.of("class2","马二","语文",97)

        };

    }

    1.6 KeyedStream

    KeyedStream用来表示根据指定的key进行分组的数据流。

    一个KeyedStream可以通过调用DataStream.keyBy()来获得。

    在KeyedStream上进行任何transformation都将转变回DataStream。

    在实现中,KeyedStream是把key的信息写入到了transformation中。

    每个event只能访问所属key的状态,其上的聚合函数可以方便地操作和保存对应key的状态。

    1.7 reduce&fold& Aggregations

    分组之后当然要对分组之后的数据也就是KeyedStream进行各种聚合操作啦(想想SQL)。

    KeyedStream → DataStream

    对于KeyedStream的聚合操作都是滚动的(rolling,在前面的状态基础上继续聚合),千万不要理解为批处理时的聚合操作(DataSet,其实也是滚动聚合,只不过他只把最后的结果给了我们)。

     

    案例1:

    public class TestReduce {

        public static void main(String[] args) throws Exception {

            final StreamExecutionEnvironment env=StreamExecutionEnvironment.getExecutionEnvironment();

            DataStream<Tuple4<String,String,String,Integer>> input=env.fromElements(TRANSCRIPT);

            KeyedStream<Tuple4<String,String,String,Integer>,Tuple> keyedStream = input.keyBy(0)

            keyedStream.reduce(new ReduceFunction<Tuple4<String, String, String, Integer>>() {

                @Override

                public Tuple4<String, String, String, Integer> reduce(Tuple4<String, String, String, Integer> value1, Tuple4<String, String, String, Integer> value2) throws Exception {

                    value1.f3+=value2.f3;

                    return value1;

                }

            }).print();

            env.execute();

        }

        public static final Tuple4[] TRANSCRIPT = new Tuple4[] {

                Tuple4.of("class1","张三","语文",100),

                Tuple4.of("class1","李四","语文",78),

                Tuple4.of("class1","王五","语文",99),

                Tuple4.of("class2","赵六","语文",81),

                Tuple4.of("class2","钱七","语文",59),

                Tuple4.of("class2","马二","语文",97)

        };

    }

    案例2:

    public class TestFold {

        public static void main(String[] args) throws Exception {

            final StreamExecutionEnvironment env=StreamExecutionEnvironment.getExecutionEnvironment();

            DataStream<Tuple4<String,String,String,Integer>> input=env.fromElements(TRANSCRIPT);

            DataStream<String> result =input.keyBy(0).fold("Start", new FoldFunction<Tuple4<String,String,String,Integer>,String>() {

              @Override

                public String fold(String accumulator, Tuple4<String, String, String, Integer> value) throws Exception {

                    return accumulator + "=" + value.f1;

                }

            });

            result.print();

            env.execute();

        }

        public static final Tuple4[] TRANSCRIPT = new Tuple4[] {

                Tuple4.of("class1","张三","语文",100),

                Tuple4.of("class1","李四","语文",78),

                Tuple4.of("class1","王五","语文",99),

                Tuple4.of("class2","赵六","语文",81),

                Tuple4.of("class2","钱七","语文",59),

                Tuple4.of("class2","马二","语文",97)

        };

    }

    1.8 Interval join

    KeyedStream,KeyedStream → DataStream

    在给定的周期内,按照指定的key对两个KeyedStream进行join操作,把符合join条件的两个event拉到一起,然后怎么处理由用户你来定义。

    key1 == key2 && e1.timestamp + lowerBound <= e2.timestamp <= e1.timestamp + upperBound

    场景:把一定时间范围内相关的分组数据拉成一个宽表

    案例:

    public class TestIntervalJoin {

        public static void main(String[] args) throws Exception {

            final StreamExecutionEnvironment env=StreamExecutionEnvironment.getExecutionEnvironment();

            env.setStreamTimeCharacteristic(TimeCharacteristic.EventTime);

            DataStream<Transcript> input1=env.fromElements(TRANSCRIPTS).assignTimestampsAndWatermarks(new AscendingTimestampExtractor<Transcript>() {

                @Override

                public long extractAscendingTimestamp(Transcript element) {

                    return element.time;

                }

            });

            DataStream<Student> input2=env.fromElements(STUDENTS).assignTimestampsAndWatermarks(new AscendingTimestampExtractor<Student>() {

                @Override

                public long extractAscendingTimestamp(Student element) {

                    return element.time;

                }

            });

            KeyedStream<Transcript,String>  keyedStream=input1.keyBy(new KeySelector<Transcript, String>() {

                @Override

                public String getKey(Transcript value) throws Exception {

                    return value.id;

                }

            });

            KeyedStream<Student,String>  otherKeyedStream=input2.keyBy(new KeySelector<Student, String>() {

                @Override

                public String getKey(Student value) throws Exception {

                    return value.id;

                }

            });

            //e1.timestamp + lowerBound <= e2.timestamp <= e1.timestamp + upperBound

            // key1 == key2 && leftTs - 2 < rightTs < leftTs + 2

            keyedStream.intervalJoin(otherKeyedStream)

                    .between(Time.milliseconds(-2), Time.milliseconds(2))

                    .upperBoundExclusive()

                    .lowerBoundExclusive()

                    .process(new ProcessJoinFunction<Transcript, Student, Tuple5<String,String,String,String,Integer>>() 

                      @Override

                        public void processElement(Transcript transcript, Student student, Context ctx, Collector<Tuple5<String, String, String, String, Integer>> out) throws Exception {

                            out.collect(Tuple5.of(transcript.id,transcript.name,student.class_,transcript.subject,transcript.score));

                        }

                    }).print();

            env.execute();

        }

        public static final Transcript[] TRANSCRIPTS = new Transcript[] {

                new Transcript("1","张三","语文",100,System.currentTimeMillis()),

                new Transcript("2","李四","语文",78,System.currentTimeMillis()),

                new Transcript("3","王五","语文",99,System.currentTimeMillis()),

                new Transcript("4","赵六","语文",81,System.currentTimeMillis()),

                new Transcript("5","钱七","语文",59,System.currentTimeMillis()),

                new Transcript("6","马二","语文",97,System.currentTimeMillis())

        };

        public static final Student[] STUDENTS = new Student[] {

                new Student("1","张三","class1",System.currentTimeMillis()),

                new Student("2","李四","class1",System.currentTimeMillis()),

                new Student("3","王五","class1",System.currentTimeMillis()),

                new Student("4","赵六","class2",System.currentTimeMillis()),

                new Student("5","钱七","class2",System.currentTimeMillis()),

                new Student("6","马二","class2",System.currentTimeMillis())

        };

        private static class Transcript{

            private String id;

            private String name;

            private String subject;

            private int score;

            private long time;

            public Transcript(String id, String name, String subject, int score, long time) {

                this.id = id;

                this.name = name;

                this.subject = subject;

                this.score = score;

                this.time = time;

            }

            public String getId() {

                return id;

            }

            public void setId(String id) {

                this.id = id;

            }

            public String getName() {

                return name;

            }

            public void setName(String name) {

                this.name = name;

            }

            public String getSubject() {

                return subject;

            }

            public void setSubject(String subject) {

                this.subject = subject;

            }

            public int getScore() {

                return score;

            }

            public void setScore(int score) {

                this.score = score;

            }

            public long getTime() {

                return time;

            }

            public void setTime(long time) {

                this.time = time;

            }

        }

        private static class Student{

            private String id;

            private String name;

            private String class_;

            private long time;

            public Student(String id, String name, String class_, long time) {

                this.id = id;

                this.name = name;

                this.class_ = class_;

                this.time = time;

            }

            public String getId() {

                return id;

            }

            public void setId(String id) {

                this.id = id;

            }

            public String getName() {

                return name;

            }

            public void setName(String name) {

                this.name = name;

            }

            public String getClass_() {

                return class_;

            }

            public void setClass_(String class_) {

                this.class_ = class_;

            }

            public long getTime() {

                return time;

            }

            public void setTime(long time) {

                this.time = time;

            }

        }

    }

    1.9 connect & union(合并流)

    connect之后生成ConnectedStreams,会对两个流的数据应用不同的处理方法,并且双流 之间可以共享状态(比如计数)。这在第一个流的输入会影响第二个流 时, 会非常有用; union 合并多个流,新的流包含所有流的数据。

    union是DataStream* → DataStream。

    connect只能连接两个流,而union可以连接多于两个流 。

    connect连接的两个流类型可以不一致,而union连接的流的类型必须一致。

    案例:

    public class TestConnect {

        public static void main(String[] args) throws Exception {

            final StreamExecutionEnvironment env=StreamExecutionEnvironment.getExecutionEnvironment(); 

            DataStream<Long> someStream = env.generateSequence(0,10);

            DataStream<String> otherStream = env.fromElements(WORDS);

            ConnectedStreams<Long, String> connectedStreams = someStream.connect(otherStream);

            DataStream<String> result=connectedStreams.flatMap(new CoFlatMapFunction<Long, String, String>() {

                @Override

                public void flatMap1(Long value, Collector<String> out) throws Exception {

                    out.collect(value.toString());

                }

                @Override

                public void flatMap2(String value, Collector<String> out) {

                    for (String word: value.split("\W+")) {

                        out.collect(word);

                    }

                }

            });

            result.print();

            env.execute();

        }

        public static final String[] WORDS = new String[] {

                "And thus the native hue of resolution",

                "Is sicklied o'er with the pale cast of thought;",

                "And enterprises of great pith and moment,",

                "With this regard, their currents turn awry,",

                "And lose the name of action.--Soft you now!",

                "The fair Ophelia!--Nymph, in thy orisons",

                "Be all my sins remember'd."

        };

    }

    1.10 CoMap, CoFlatMap

    跟map and flatMap类似,只不过作用在ConnectedStreams上

    ConnectedStreams → DataStream

    1.11 split & select(拆分流)

    split

    1.DataStream → SplitStream

    2.按照指定标准将指定的DataStream拆分成多个流用SplitStream来表示

    select

    1.SplitStream → DataStream

    2.跟split搭配使用,从SplitStream中选择一个或多个流

    案例:

    public class TestSplitAndSelect {

        public static void main(String[] args) throws Exception {

            final StreamExecutionEnvironment env=StreamExecutionEnvironment.getExecutionEnvironment();

            DataStream<Long> input=env.generateSequence(0,10);

            SplitStream<Long> splitStream = input.split(new OutputSelector<Long>() {

                @Override

                public Iterable<String> select(Long value) {

                    List<String> output = new ArrayList<String>();

                    if (value % 2 == 0) {

                        output.add("even");

                    }

                    else {

                        output.add("odd");

                    }

                    return output;

                }

            });

            //splitStream.print();

            DataStream<Long> even = splitStream.select("even");

            DataStream<Long> odd = splitStream.select("odd");

            DataStream<Long> all = splitStream.select("even","odd");

            //even.print();

            odd.print();

            //all.print();

            env.execute();

        }

    }

    1.12 project

    含义:从Tuple中选择属性的子集

    限制:

    1.仅限event数据类型为Tuple的DataStream

    2.仅限Java API

    使用场景:

    ETL时删减计算过程中不需要的字段

    案例:

    public class TestProject {

        public static void main(String[] args) throws Exception {

            final StreamExecutionEnvironment env=StreamExecutionEnvironment.getExecutionEnvironment();

            DataStreamSource<Tuple4<String,String,String,Integer>> input=env.fromElements(TRANSCRIPT);

            DataStream<Tuple2<String, Integer>> out = input.project(1,3);

            out.print();

            env.execute();

        }

        public static final Tuple4[] TRANSCRIPT = new Tuple4[] {

                Tuple4.of("class1","张三","语文",100),

                Tuple4.of("class1","李四","语文",78),

                Tuple4.of("class1","王五","语文",99),

                Tuple4.of("class2","赵六","语文",81),

                Tuple4.of("class2","钱七","语文",59),

                Tuple4.of("class2","马二","语文",97)

        };

    }

    1.13 assignTimestampsAndWatermarks

    含义:提取记录中的时间戳作为Event time,主要在window操作中发挥作用,不设置默认就是ProcessingTim

    限制:

    只有基于event time构建window时才起作用

    使用场景:

    当你需要使用event time来创建window时,用来指定如何获取event的时间戳

    案例:讲到window时再说

    1.14 window相关Operators

    放在讲解完Event Time之后在细讲

    构建window

    1.window

    2.windowAl

    window上的操作

    1.Window ApplyWindow Reduce

    2.Window Fold

    3.Aggregations on windows(sum、min、max、minBy、maxBy)

    4.Window Join

    5.Window CoGroup

    2. 物理分区

    2.1回顾 Streaming DataFlow

     

    2.2并行化DataFlow

    2.3算子间数据传递模式

    One-to-one streams

    保持元素的分区和顺序

    Redistributing streams

    1.改变流的分区

    2.重新分区策略取决于使用的算子

    a)keyBy() (re-partitions by hashing the key) 

    b)broadcast()

    c)rebalance() (which re-partitions randomly)

    2.4物理分区

    能够对分区在物理上进行改变的算子如下图所示:

     

    上面算子都是Transformation,只是改变了分区。它们都是DataStream → DataStream。

    2.5 rescale

    通过轮询调度将元素从上游的task一个子集发送到下游task的一个子集。

    原理:

    第一个task并行度为2,第二个task并行度为6,第三个task并行度为2。从第一个task到第二个task,Src的子集Src1 和 Map的子集Map1,2,3对应起来,Src1会以轮询调度的方式分别向Map1,2,3发送记录。从第二个task到第三个task,Map的子集1,2,3对应Sink的子集1,这三个流的元素只会发送到Sink1。假设我们每个TaskManager有三个Slot,并且我们开了SlotSharingGroup,那么通过rescale,所有的数据传输都在一个TaskManager内,不需要通过网络。

     

    2.6任务链和资源组相关操作

    startNewChain()表示从这个操作开始,新启一个新的chain。

    someStream.filter(...).map(...).startNewChain().map(...)

    如上一段操作,表示从map()方法开始,新启一个新的chain。

    如果禁用任务链可以调用disableChaining()方法。

    如果想单独设置一个SharingGroup,可以调用slotSharingGroup("name")方法。

  • 相关阅读:
    python(4):Python应用场景
    python(3):python的优缺点
    MySQL(17):用户登录实例
    MySQL(16):参数化、封装
    MySQL(15):与python交互的增删改查
    MySQL(14):与python交互的类型
    MySQL(13):索引
    c++ --程序编译
    c++ -- 顺序容器操作
    c++ --string操作
  • 原文地址:https://www.cnblogs.com/dajiangtai/p/10711679.html
Copyright © 2011-2022 走看看